Find care now
If you are experiencing a medical emergency, please call 911 or seek care at an emergency room.
MedStar Health Researchers demonstrate that natural language processing can help break the cycle of violence, closing gaps in screening and reducing the need for revisits to the emergency department.
About 25% of women and 10% of men in the U.S. have experienced physical or sexual intimate partner violence (IPV) or stalking. Abuse often leads to multiple hospital visits to treat physical injuries, while mental health and socioeconomic challenges may go unidentified and untended.
Similarly, violence in the workplace is common in healthcare with 34.4% of U.S. healthcare workers having reported physical or verbal violence. Our new study in natural language processing technology may reveal a key to breaking the cycle for both IPV and workplace violence (WPV).
At the point of emergency care, it can be challenging to pinpoint the situation that caused an injury. Screening questions may go unanswered, and proper ICD-10 codes may not be entered into the patient’s medical records to indicate the role of violence. Without those notes, patients may not get referred for much-needed mental health and socioeconomic care to break the cycle of violence.
Data from our study shows that analyzing free-text data for words and phrases that could indicate IPV with natural language processing technology can help identify violence. This technique can be used at home and in the workplace, giving healthcare providers context they need to refer patients for holistic care and health systems administrators the data they need to safeguard providers and staff.
Identifying intimate partner violence in health records.
As part of my PhD thesis, I collaborated with emergency medicine, public health, and surgery departments at the Emory University School of Medicine in Atlanta, GA.
We developed a natural language processing algorithm to identify patients exposed to IPV by examining anonymized patient medical records with specific ICD-10 codes. We also examined relevant research to list known terms indicating IPV, such as “domestic violence” or “attacked at home.”
Our team fine-tuned the algorithm with this terminology, resulting in a set of free-text terms we could use to identify IPV not noted with an ICD-10 code or answer to a screening question.
Only about 30% of the cases our algorithm identified as IPV had an ICD-10 code to match, indicating these codes for violence are underused, and many patients who are victims of IPV may go unrecognized.
The algorithm allows us to better understand our patients’ challenges and provide them with support services, such as referrals for mental health care, housing, food, and childcare.
More collaboration with specialists in IPV care is needed before this tool is ready to use in the emergency department. In the meantime, the natural language processing concepts we studied and tested are being researched at MedStar Health to help understand and reduce the impacts of workplace violence in the healthcare setting.
Based on our work in IPV, we received an MHRI Early Investigator Award in August of 2023 to apply natural language processing techniques to examine workplace violence in the emergency department and inpatient setting at MedStar Health.
Our study uses a similar process to examine patient safety event reports and security office statements. Much like in IPV, free-text analysis can be applied these reports to identify patterns of violence and integrated into patient records to help keep providers safe.
For instance, if a patient has a history of abuse toward female providers, their record could indicate this pattern and recommend a male provider and or additional security measures. Identification is the first step toward the prevention of future incidents.
Our machine-learning technology offers an opportunity to make a difference in patients’ lives, make workplaces safer for everyone, and put a stop to the cycle of violence.
